Lately, I've been hearing everyone praise AI Agents for being fully automated on the chain, but I still hold that one belief: just because it can run doesn't mean it can cover all risks. If I really let it do the work, there are at least a few steps I still need to oversee: first, check if its permissions are being misused (I’ll manually revoke unlimited permissions myself), then verify if the contracts it interacts with are among the ones I’ve marked, and I’ll still review the amount and recipient in the signature popup—I'd rather be slow. Also, cross-chain swaps and pool changes are very unpredictable because slippage and routing can change suddenly; an Agent might think it’s “more optimal,” but I just see it as “more dangerous.” Recently, during the testnet points farming phase, it was the same—no one knows if the mainnet will actually issue tokens, and the Agent was faster than anyone else. Anyway, I see the interactions starting to look like scripted volume farming, and when on-chain tags turn red, I’ll just withdraw first and ask later… I’m just an ordinary user with a bit of paranoia; living longer is more important than anything.